Skip to content

Security: bohselecta/chat-said

Security

docs/SECURITY.md

Security and Privacy

Intended boundary

This build is designed for a trusted home/studio LAN. It is not configured for direct exposure to the public internet.

Implemented controls

  • Client pairing code and random persistent tokens
  • Same-origin API and artifact access
  • No cloud service requirement
  • Original audio not returned by agent search tools
  • Agent write disabled by default
  • Agent writes confined to agent-output/
  • Path sanitization
  • Request-size limits
  • SQLite foreign keys and WAL
  • Atomic canonical-file writes
  • Audit events for project, queue, revision, artifact, pairing, engine, export, wake, and agent actions

Production hardening work

Before internet exposure or regulated clinical deployment:

  • terminate TLS with a trusted certificate;
  • add device-token revocation UI and rotation;
  • encrypt sensitive data at rest;
  • protect backups;
  • add automatic locking and OS account separation;
  • formalize retention/deletion policy;
  • conduct threat modeling and external review;
  • add consent and recording-law workflows;
  • validate all applicable healthcare and professional obligations.

Local processing reduces data exposure; it does not automatically create regulatory compliance.

There aren't any published security advisories